Version history


  • Separate function: generate / preprocess works only if each tag turns on in input.yaml.
  • Bug fix: Inconsistent position from ASE.
  • Memory issue fix: pair_nn.cpp for more atom type without memory leak.


  • Replica ensemble: you can estimate the standard deviation of atomic energy to detect the atom out of training set.
  • Bug fix: LAMMPS atom type matching error, memory allocation


  • use_stress: (default false) if you set use_stress true in neural_network of your input.yaml , you can train your neural network with virial stress tensor
  • LAMMPS Optimization: Using optimization in for loop and applying ‘memoization’, it becomes around 2 times faster!